If you have handed your bank login to an AI assistant so it can track spending, pay bills or move money between accounts, your bank would like a word. Actually, it would like you to sign something, and a lot of people already have without noticing.
This week Citibank, the third largest commercial bank in the US, updated its Online User Agreement to say that customers who give a third-party AI agent access to their accounts are responsible for whatever that agent does. Then it went a step further than anyone before it: if the AI tool leaks your credentials or gets hacked, you are also responsible for what the hackers do with your account. PNC changed its terms the same day, and Citizens emailed customers with a similar warning.
They join JPMorgan Chase, Bank of America and Wells Fargo, which moved in the same direction over the past four weeks. Put together, the biggest names in American banking have now drawn the same line: AI agents are allowed to look, maybe, but if one of them touches your money, it is on you.
The short version
- Who: Citi, PNC, Citizens, JPMorgan Chase, Bank of America and Wells Fargo
- What: anything an outside AI agent does with your login counts as something you did
- New at Citi: that includes losses after the AI tool itself is hacked or leaks your password
- The exception: AI tools the bank provides itself are not covered by the extra liability
- Your safer option: read-only connections that never get your password
How we got here in four weeks
None of this came out of nowhere. Banks started noticing software, not people, logging into customer accounts this fall. On October 22, Axios reported that several large banks had detected agent activity they had not approved, with some agents calling customer service lines or trying to do more than they were asked. Bankers described the situation as “horrifying.” Morgan Stanley disputed that rogue agents had reached its platforms, and Reuters said it could not independently verify the findings.
Bank of America’s response on October 27 was almost polite. An email titled “Protecting your finances in the AI World” asked customers never to share their username and password with anyone, “including an AI-enabled tool or agent that offers to act on your behalf.” It ended with the sentence that is now spreading across the industry: if you do share, “you’re responsible for activity resulting from that access.”
What each bank now says
The wording differs, but the direction is the same everywhere. Here is where the six banks stand, based on their published terms, spokesperson statements and reporting from TechCrunch and others.
| Bank | What changed | The detail to know |
|---|---|---|
| Citi | Online User Agreement updated this week | You are liable for agent actions and for what hackers do if the AI tool is breached |
| PNC | Terms changed November 18 | Same direction as Citi; full wording not yet widely reported |
| Citizens | Customer email this week | You are on the hook for anything your AI assistant does |
| JPMorgan Chase | Blocked outside agents, revised digital terms | Instructions sent with your credentials count as proof you authorized them |
| Bank of America | Warning email, limits on third-party access | Restrictions reach some services, including Zelle |
| Wells Fargo | New Section 15 in its Online Access Agreement, effective Nov 24 | You must keep proof that you revoked an agent’s access |
Wells Fargo’s version is the most detailed so far, and it reads like a preview of where everyone else is going. Its new section defines an AI agent as any software that acts on your behalf, whether autonomously or not. Any instruction an agent sends with your login is treated as coming directly from you, “without further verification.” You are responsible for unintended actions, for mistakes caused by the agent misreading your instructions, for actions based on manipulated inputs, and for cases where the agent goes beyond what you allowed. The bank says it does not monitor agent activity at all. If you cut an agent off, you have to keep evidence and hand it over if asked, and you stay responsible until you tell the bank.
“Customers who choose to grant account access to third-party AI agents are responsible for activities conducted through those agents.”
Citi spokesperson, to TechCrunch
“Actions based on manipulated inputs” is the scary phrase
That line in the Wells Fargo terms deserves a second look, because it describes a real attack. AI agents read web pages, emails and documents while they work. Someone can hide instructions in that content, a trick known as prompt injection, and steer the agent into doing something you never asked for, like paying a fake invoice. Under these terms, that payment is treated as yours.
It is not a hypothetical worry. We have seen how hard it is to keep agents on a leash, including when California said OpenAI lost control of its AI agents 80 times. Banks looking at that record have decided they do not want to be the ones paying for the next incident.
Doesn’t federal law protect me?
Partly, and this is where it gets murky. Every one of these terms says it applies “subject to applicable law,” and the most important law here is the Electronic Fund Transfer Act and its rule, Regulation E. It caps what you lose when someone makes an electronic transfer from your account without permission.
| When you report an unauthorized transfer | Most you can lose under Regulation E |
|---|---|
| Within 2 business days of learning your login or card was compromised | $50 |
| After 2 business days | Up to $500 |
| More than 60 days after your statement shows it | Potentially unlimited for later transfers |
The catch is the word “unauthorized.” Regulation E says a transfer is not unauthorized if it was made by someone you gave your access device to, until you tell the bank that person is no longer allowed. Banks are betting that an AI agent you handed your password to falls into exactly that bucket. The rule was written decades before autonomous software existed, and no court or regulator has ruled yet on how it applies to an agent that goes off script. Until someone does, the bank’s reading is the one that shapes what happens when you call to dispute a charge.
Worth knowing: the hack scenario in Citi’s terms is the hardest one to argue with. If a breach at the AI company exposes your password and a criminal drains your account, the bank’s position is that the leak started when you gave the password away. A credential you never shared cannot be leaked by someone else’s app.
It’s also about who owns the customer
Security is the official reason, and it is a real one. But there is a business angle too. AI tools that scan your accounts can point out fees you are paying for nothing or nudge you away from a savings account with a tiny interest rate. That is money banks would rather keep. Notice that Citi’s terms give a pass to AI agents that Citi itself provides. If agents become the main way people manage money, banks want to be the ones running them.
The same pattern is playing out outside finance. United Airlines now restricts personal AI agents from booking flights, and retailers are split, as we saw when Walmart gave shopping agents their own profile while Amazon took them to court.
How to keep using AI with your money, safely
You do not have to give up on AI budgeting tools. You just have to stop handing over the keys. The difference between the two approaches is what decides who pays when something breaks.
| Approach | Can it move money? | Who carries the risk |
|---|---|---|
| Giving an agent your username and password | Yes, anything you can do | You, under the new terms |
| Read-only link through an approved aggregator | No | Much lower, since no password is shared |
| Login brokered by a password manager | Depends on the task | You approve each login; check bank terms |
| Your bank’s own AI assistant | Within the bank’s limits | Not covered by the extra liability at Citi |
- Prefer read-only. Anthropic’s new tool for Claude is a good example: it connects through Plaid and can read your accounts but cannot move a dollar. That is the safest shape for this kind of product.
- Never paste your bank password into a chatbot or agent. If a tool asks for it directly, treat that as a red flag.
- Check what you have already connected. Review linked apps in your bank’s security settings and revoke anything you do not use. Screenshot the confirmation, especially if you bank with Wells Fargo.
- Change your password if you ever shared it. Revoking an app does not help if the tool still remembers your login.
- Turn on transaction alerts. The faster you spot a strange transfer, the stronger your position, whatever the fine print says.
AI companies keep promising that agents will handle the boring parts of life for us, and banking is near the top of everyone’s list. The banks have now made their position clear. Until the law catches up, an agent with your password is not your assistant in their eyes. It is you.
About this article: GeekBlog covers U.S. technology news, AI, phones, smartwatches and gaming. Every story is written and checked under our Editorial Policy. Spotted a mistake or have a story tip? Contact our editors.

